The Economics behind the Everpix Shutdown Decision

tl;dr

Everpix sold its product at a marginal loss and closed its doors after the financing ran out. Since the marginal costs always exceeded the marginal revenue we now know that Everpix should have shut its doors immediately as it never could be a viable business in either the short or long runs.  There doesn't appear to be a what-if cost structure change that it could have made realistically to stay in business.  Shutting down was the right decision for the business, and this evidence suggests it should have shut down a long time ago.

What is Everpix and Shutdown Summary

Everpix was a paid photo sharing and storage product that attempted to collect and organize all photos taken by consumers into one online service.  It began in 2011 and shut down in November 2013.  The Verge had the most complete reporting on the shutdown.  In January 2014 the Everpix team released many of its internal financials publicly on Github.  This is among the first public releases of data I can recall of a failed startup, yet the analysis by the team was incomplete.

Context for this Analysis and my Bias

Twelve months ago (January 2013) I began working on a paid photo sharing project that would have been a competitor to Everpix. I knew about Everpix, as well as rumors that Flickr was getting a redesign, but my team and I thought we could build a viable product in stealth mode that would win on features. Three months later we had a finished product with a soft launch (web site and iOS app), including three customers sharing their wedding photos via our product. In April we abandoned our product without a public launch on the hunch that the major competitors were selling their products at losses and we couldn’t compete on price. We modeled the economics of hosting and storing user photos and concluded that the prevailing price (~$5-10/month) was 1/4 to 1/6 of the variable cost using retail AWS and Rackspace prices. It seemed only players which owned significant infrastructure (Yahoo!, Google, Facebook, Apple, Amazon, and Microsoft) could operate profitably. It also gave us pause that major AWS customers like Dropbox who get >50% discounts were also likely selling at losses for some customer segments.

We predicted that Everpix and a few other would-be competitors would shut down by the end of 2014, we sold the technology to a major player (under NDA), and we’ve since shifted to another stealth mode startup. However, until Everpix released its financials last week, we were only speculating about their internal costs. Now we can evaluate how accurate our predictions in April 2013 were. Since April a few others have noticed as well.

A bit about me: I have a formal education in computer science, economics, and accounting from a major US university. In my professional life I’ve worked as an engineer at a few large software companies, as a CTO of two startups, and I’ve freelanced as a due diligence consultant for VC firms and angel investors. Because I am deeply technical in both how technology works (I write Java, Scala, Objective-C, and JavaScript) and how businesses work (e.g. economics and accounting), I am conveniently positioned to evaluate and predict holistically the potential of different software businesses. These were the skills I used to predict shutting down our photo venture and the photo sharing market irregularities, which in retrospect, proved accurate.

Part 1: Why did Everpix fail?

Everpix was not profitable, either overall or operationally. The following graph shows the total reported revenue, costs, and financing rounds for the company:

You see how the total reported revenue $200k (blue) was 1/10 of the reported costs $2M (red). The only way this company operated at all was by the 3 financing rounds totaling around $2M (purple).

One challenge with a simple summary graph like this is all companies have initial periods of time that look this way. Everyone starts out using some form of financing to pay for the supermajority of costs. Some companies use financing for the first few months and others for the first year. The trick is to understand the cost structure and potential for revenue to grow to see if it is possible for the revenue to cover costs ever. In economics we generally codify this using a simple rule: Keep producing products so long as the marginal revenue is greater than or equal to the marginal cost.

What revenue potential did this market have?

One challenge with the photo market is that revenue growth is constrained by three large forces (in the US market, at least):
  1. Good-enough solutions like Facebook, Instagram, Free Flickr, Twitpic, Google+, and Apple iCloud charge zero revenue.  You could argue the price of privacy is non-zero, but it's unlikely to be more than $10/month.
  2. The prevailing price for premium consumer products is around $10/month due to multiple alternatives (Dropbox, Flickr Pro, SmugMug, iCloud extra storage, etc.).
  3. There are non-service alternatives that charge $2-$10/month depending on amortization schedules (i.e. a hard drive that lasts for 1-3 years costs only $100 or so dollars today).
Because of these constraints, selling a standalone product must be in the $10/month range to be competitive. I interpret startups as price-takers in the photo sharing market, which means that startups must control costs heavily in order to be profitable.

Everpix’s primary revenue came from subscription revenue, which was in the $10/month range as expected. So let’s look at costs.

What were Everpix’s cost breakdown?

As I discussed in the previous section, I believe that Everpix was a price taker in the photo sharing market. Therefore, in order for Everpix to be successful it had to have costs less than the prevailing price in the market.

First, let’s break down Everpix’s costs into broad categories:

Visually and numerically, the largest expense Everpix has are on its people: Personnel, Consultants, Legal, PR, and accountants represent 70% of all expenses. Wow! One anecdotal note: Most expenses I’ve seen at startups I evaluate are on its people.

The next question is whether spending 70% of costs on people is economically sustainable for this business. There is nothing inherently wrong with 70% of costs going to people. Some businesses, for example, spend nearly 90% of costs on people (maids) and other businesses spend only 15% (paperclip manufacturing). They all can be sustainable. It just depends on the business. My suspicion, however, is the Everpix spending was reckless, but that label -- reckless -- is based on personal preference and cannot at this point be defended with analysis.

The way to answer whether Everpix is sustainable is to remember microeconomics 101: Businesses have fixed and variable costs. Variable costs are defined as costs incurred as a function of the supplied output. Fixed costs do not have supplied output as an input variable. Expressed as a mathematical expression:

cost(x)=variable(x) +fixed

Since Everpix was a software startup, when customers used the product (uploading, storing, and displaying photos) Everpix incurred costs related to the server processing, storage, and bandwidth only. The cost to write the software did not vary significantly based on the number of users. Developing the software (their personnel costs) and all the other overhead (legal, PR, consultants), therefore, are fixed costs for Everpix.

To simplify the analysis, we break down Everpix’s costs using only the Variable and Fixed classifications:

In the next section we investigate the variable costs to understand if the Everpix business was sustainable.

Part 2: Was closing Everpix the right economic decision?

In economics we say that a business should supply a product if the marginal revenue is greater than the marginal cost. One outcome of this rule is that the fixed cost a business faces is generally irrelevant, and if a business has positive margins it will be sustainable generally in the short run. The reason is that businesses generally can find financing necessary to pay for fixed costs. The trick becomes in calculating the amortization schedule for the fixed costs that should be included in the marginal cost calculation, which in many industries is more art than science and helps inform interest rates over the long term.

In the case of Everpix as a software business, my experience has shown an extreme appetite for investors to finance fixed costs on the promise of “user growth to be part of an acquisition.” I would expect no different here.

So, to answer the sustainability question let’s begin by calculating the marginal cost and revenue without amortizing the fixed costs. We then will apply different theories of amortization schedules to determine whether Everpix should have closed in the long run.

Breaking down the variable costs

In the previous section we investigated the overall cost breakdown of Everpix and discovered that 20% of the costs were variable (aka operational in accounting lingo). One of the fantastic resources included in the Everpix dataset is the raw AWS billing information. This data is priceless and accounts for almost all of the variable (operational) costs!

I went through and grouped the different costs into 4 categories:
  1. Bandwidth
  2. Storage
  3. Servers (Processing)
  4. Database (RDS)

I also like looking at these data as sparklines to see how the team made changes to control costs for each type:


The dip in the middle of 2013 in database and server costs were likely due to deliberate cost cutting by the team since I can’t find specific AWS price cuts. If I were brought in to do due diligence I would plan to ask the team about significant development milestones during 2013. The team’s first answer would focus on features, and I would plan to follow up with the specific question about what happened in May and June that caused the database costs to decrease. You’d be surprised about how much higher quality of answers you get after you let the team parade out their pre-planned bullet points first.

A few other observations:
  1. RDS as a database is expensive. About 1/3 of the operational costs were due to the database. When I see this choice made by startups, it comes across as lazy and immature. RDS is great for enterprises that have legacy schemas but have a CIO mandate to “use the cloud.” In my humble opinion the Everpix team would have been better off running its own databases and sharding accordingly. Building their own would have been higher fixed costs (development time) but yielded lower variable costs.
  2. The team had a variety of EC2 instances running for availability. These servers accounted for 1/3 of the operational costs. All of these instances seemed to be paid for using on-demand rates. I didn’t look more closely, but I think there were opportunities to save on server costs with more spot instances and pre-paying.
  3. Why doesn’t bandwidth trend along with any of the other metrics? I would have expected it to follow storage, visits, or trailing visits. One theory is that in the beginning the team was backing up photos outside of AWS and then stopped. Another is that early customers tried and left the service, possibly after Flickr announced its redesign. I would also want to probe this non-trend but I wouldn’t expect to find anything.

Calculating the Accrual-based Revenue

Now that we have the monthly costs we next need to calculate the monthly revenue. One challenge with Everpix is that the company sold three offerings: monthly via Apple, monthly via its site, and yearly pre-paid via its site. In order to calculate the monthly margins on the way to calculating marginal revenue and marginal costs, we need to convert all revenue from the cash basis reported by Everpix to an appropriate accrual basis.

In general in modern accounting you can only recognize (or count) revenue when you render the service or sell the product. So when customers pre-pay for a year, you can only recognize 1/12 of the cash they give you as revenue each month. This matching is called an “accrual” basis and contrasts to an often deceiving “cash” basis. I went through and did straight-line (or linear) recognition of yearly subscriptions to calculate the monthly revenue:


We see a general trend of up-and-to-the-right. It’s not quite a hockey stick, but the revenue and growth look positive. Expect to see this graph plastered everywhere by Everpix because it must mean the business is healthy! The next step is to combine the revenue and costs to infer the gross income from operations.

Calculating the Gross Operating Income

Now that we have the costs from operations (variable costs) and the revenue from operations (accrual-based monthly revenue), we now can calculate the gross income for each month for Everpix. This result will tell us how sustainable the core of the Everpix business is.

Drum roll…..

This is not a good picture. In fact, this is the picture that likely prevented anyone else from investing in Everpix. The graph shows that every month Everpix was selling its product for an operating loss (negative income!). This is the definition of unsustainable. Remember, this graph is based on the monthly accrual revenue less the variable (operating) costs incurred; it is the core of the business model and represents the short run.

There are a few serious issues with the trends in this graph:
  1. This is a paid product, yet every month is negative. And not just a little negative, but in percentage terms the gross margin is a range of -20% to -150% of revenue.
  2. The trend of the gross operating income does not improve.
  3. We have not yet amortized any of the significant fixed costs yet, which means the net loss per month is even worse.
  4. Looking at the metrics released, I am skeptical the founders ever knew the gross margin was this negative. They likely knew costs were high but they didn’t understand why their business would fail.
It is no surprise that Everpix shut its doors. The business was a failure, and it cannot operate even in an open source model. There isn’t even a point in amortizing the fixed costs since the variable costs are negative. Had we seen this graph during any of the financing rounds, we would have known the company would fail when the financing ran out. That’s exactly what happened.

Were the losses improving?

Now, the microeconomic rule I quoted above was for the marginal cost and marginal revenue. So far we’ve calculated the gross costs and gross revenue by month. Calculating marginal costs is a bit tricky since we don’t know the distribution of users over time. But we can infer direction of the value (i.e. whether it is positive or negative) by looking at the averages. The following graph is the average cost versus average revenue per user for Everpix by month:

This picture is even worse than the gross operating income. This is the smoking gun. It shows that for every month the average operating cost exceeded the average revenue per user without hope of changing.  That means that the marginal costs likely exceeded the marginal revenue for most users. Again, we’d have to know the specific distributions (how many photos per user, for example) to know how many users generated negative gross margins, but I suspect at least half of the users dragged the business down. Users should be building a business up!  For the business to be sustainable the revenue area (blue) has to exceed the cost area (red).

We do see the impact of the May-June 2013 cost cutting on servers and databases in this graph where the average operating cost approaches the average revenue.  However, this graph reveals the last bit of insights about the business model:
  1. Everpix needs to double prices or cut variable costs in half in order to break even operationally in the short run. Given the market dynamics, it seems impossible to double prices to $20/month. Also, a 50% discount on AWS is typically reserved is for a Dropbox-scale service. Everpix would need some killer connections to pull it off.
  2. (Back of the napkin math) Everpix needs to quadruple or quintuple prices to begin to cover its fixed costs in the long run. This seems even less likely than doubling prices given the market.
  3. The revenue per user is flat. Outside of what looks like a price increase or better conversions in March, it does not look like there was any upsell for users once they bought a subscription.
The losses were not improving and there was no hope of increased per-user revenue growth.

Could Everpix have changed its cost structure to survive?

Let's ask a couple what-ifs to see if there was any way to save the business.

First, what if the company had removed RDS entirely?  Would reversing that technology decision had been enough to save the product?  Below is the same average revenue vs average cost excluding RDS:

Again, we see that at no point does average revenue exceed the average cost.  The final month's numbers include the impact of the shutdown notice, so there is a false trend of hope at the very end.  While I think using RDS was foolish, it wasn't the proverbial straw that broke the camel's back.

Second, what if the company had been able to make the servers free?  What if they only had to incur the AWS storage and bandwidth costs?
We see for the first time hope of a sustainable business!  In percentage terms the average revenue has a gross margin of 30% over the storage costs.  Thirty percent is a respectable margin, but unfortunately I don't know of any way to run a software business and only incur storage costs.  Perhaps if Everpix had connections they could have negotiated for better deals, but the sum total of the deals would have needed to shed all of the server and RDS costs (> 50%) to find positive gross margins.  I can imagine AWS offering these deals to major partners like Dropbox, but Everpix's scale was too small to command such consideration.  Shedding the necessary costs to gain positive margins simply isn't realistic.

Since the marginal costs always exceeded the marginal revenue we now know that Everpix should have shut its doors immediately as it never could be a viable business in either the short or long runs.  There doesn't appear to be a what-if cost structure change that it could have made realistically to stay in business.  Shutting down was the right decision for the business, and this evidence suggests it should have shut down a long time ago.

Ending on a positive note: The metrics Everpix showed off

Now that we know the Everpix business had no future, I want to revisit the graphs that the company showed off to illustrate one of the unfortunate trends in the startup world: Growth != Sustainability. Revenue, Costs, and Profits -- good, old-fashioned business rules -- are still important.

Contrast the team’s message of positive growth (all of these graphs are up and to the right) with what know are grossly negative margins:


Next time you see an article about growth or a startup talking about users, remember to consider the underlying costs and revenue at play. Ask yourself: Are they actually making a positive margin on each sale? If it s a free product then they likely are hoping to get acquired since another dirty secret is advertising doesn’t pay that well. If it is a paid product and the margins are negative, run the other way.

Also, notice the sales volume graph is cash-basis!  Super deceiving!

Part 3: Lessons to Takeaway

Vainly, my team and I were right about the photo sharing business. At least one incumbent was selling at a loss, and I think we were just as right about the broader trend. While we estimated the prices were 1/4 to 1/6 of variable costs, in the case of Everpix the price was closer to 1/2 of variable cost. We were right on direction, but wrong on magnitude by 50%. We feel vindicated to have switched to another -- profitable -- business. In contrast, I would remain skeptical about the team behind the Everpix product. They may be very good at building software and products, but I am extremely concerned by their ability to build a sustainable business.

Takeaway 1: Applying simple microeconomic rules can be done with basic data and should be done early

The first important takeaway is we did this entire analysis without looking at features or product market fit. We looked at trailing data and applied simple microeconomic rules to determine if the product was viable.  I would recommend doing this simple pro forma analysis as early as possible during your product development.  Especially with AWS you can estimate usage using only a few months of data.  It seems obvious, but many trends in the software startup world ignore this advice. Some examples include: Hackathons, Startup Weekend, “Growth Hacking,” and parts of the hit-based VC culture.

Takeaway 2: Details matter, especially with variable costs

The second important takeaway is that the details matter. I bet the Everpix team looked at the costs all up on the P&L sheet, they saw that AWS was something “small” like 15% of costs and didn’t dive in to understand. This isn’t to say that all details matter, but when it comes to variable costs -- the core part that makes a business sustainable -- you need to know that you are selling your product for marginal gain. It’s clear that Everpix at a minimum didn’t know and possibly ignored this important detail.

For example, picking RDS is the single most suspicious engineering decision I saw without looking at the code or designs. They picked a very high variable-cost database when it should have been clear from the beginning how thin their margins were. I bet if I saw the actual code I would notice poor time-space tradeoffs that properly balanced user experience with variable costs (i.e. how many different resolutions did they store? Could they get away with computing on the fly?). I am deeply grateful to be able to look at this data and must thank the Everpix team profusely for sharing publicly, but every bit of my experience says the team is deeply immature and inexperienced to build a world-class business. Selling paid products at a marginal loss is unforgivable to me.

Takeaway 3: Unfortunately Everpix is illustrative of VC-funded startups on the west coast

A third important takeaway is that this company is illustrative of many VC-funded startups that I’ve seen up close. Don't walk away from this analysis thinking the problem and myopia are limited to this one company.  They are not.  I’ve asked many a founder about profitability, and you’d be surprised how frequently founders are unaware of their positions. “We’re bringing in cash.” “We were profitable from the first day.” “We’re selling advertising.” This lack of profitability is a consistent theme at YC, 500 Startups, A16Z, and TechStars. If the VC bubble pops, it will be because of inattention to profitability.  "We'll just make it up on volume user growth..."

Takeaway 4: Be skeptical of the founders' and press narratives

While The Verge reported on the drama surrounding an upcoming AWS bill, the founders attributed their failure to everything except the negative margins:
  1. "They spent too much time on the product and not enough time on growth and distribution."
  2. "The first pitch deck they put together for investors was mediocre."
  3. "They began marketing too late."
  4. "They failed to effectively position themselves against giants like Apple and Google, who offer fairly robust — and mostly free — Everpix alternatives."
  5. "And while the product wasn't particularly difficult to use, it did have a learning curve and required a commitment to entrust an unknown startup with your life's memories — a hard sell that Everpix never got around to making much easier."
Some of the top rated HN comments presumed, "they had a functional business model, as they have specifically said that their user revenues outpaced their customer acquisition and service costs. They had positive customer lifetime value. The issue was that they couldn't acquire enough customers to pay their overhead."  We've seen this is not the case.

And one of the Everpix employees even said "AWS infrastructures costs were already being covered by subscription income."  Based on the numbers released that is not true.  Either the numbers released are wrong, I misunderstand the numbers released, or the employee wasn't aware of the underlying cause.  I suspect it was the latter since my totals (total costs, total revenue, etc) match the publicly reported numbers.

So, as far as I can tell, outside of a few kooks on HN no one has said out loud what should now be obvious: Everpix was selling its product at a loss and hoping to make it up on volume.  No change to the pitch decks, marketing, product features, or timing could have saved them.

Part 4: Final Thoughts

These are opinions that I can't defend with data but strike me as the advice I would give to other startups in a similar position.

Wrong team size and fit

I believe that the company had too many people. Picking the right number and right mix of talent is very much an art, but I think the 5 full-time employees and 3 founders was twice as many as the company needed. I would have built the team in this way:
  1. 4 Full time, including founders:
    1. Senior backend engineer that handles API, processing, storage and cost control (dev ops).
    2. Senior frontend engineer that makes a great UI experience. Also capable of one other role (Windows, Android, backend, or managing contractors)
    3. Junior or Senior Apple developer (for both Mac and iOS)
    4. Optional: Junior or Senior domain expert for credibility that also does one other role such as customer support and QA
  2. Part time:
    1. Contract out Android app and Windows app as needed
    2. Contract out design as needed

Reckless spending

"Reckless" is a strong word to throw around, but in this case it feels apt.  There was no way for the company to be profitable, and the founders should have known this.  They had a responsibility to know this.

Paying the best PR firm in the world to sell more products at a loss will not turn the company around and used to be outright fraud just 30 years ago.  I would be more forgiving had I seen the company control its costs more consistently. But, they didn’t, and since I haven’t seen mention by any of the founders of a failure of the core business model I think they don’t know or are being disingenuous about the outcome.  The same goes for the expensive office rent, the nice Apple computers they bought themselves, the travel, and recruiting fees.  Everpix was not a viable ongoing concern and responsible officers would have wound it up long before spending every penny of financing.

Similarly, it's hard to take the Seed and Bridge Loan investors seriously since it should have been abundantly obvious this business could not survive. Any simple pro forma would have revealed the negative margins, and those negative margins should have put everyone off. How do you defend an investment in this business outside of a hail marry acquisition?  It looks like the Path executive team got spooked during due diligence even though their product desperately needs 10k paying customers.

Didn't Mattermark...
I also remember Mattermark ranking Everpix highly, but alas I can't find any evidence to back this claim up.  Chalk up this poor memory as wild speculation, but if you can find a link or a cached report that was bullish on Everpix from my "favorite" researchers, please send it my way.  Be wary of the siren of companies that sell raw data... statistics can be used to form a variety of narratives.

FAQs about Growth

Here are some questions I've heard at cocktail parties this weekend:
  • "Didn't Everpix just need to grow more?"
  • "Wouldn't Everpix have been profitable at a larger scale?"
No.  Each additional customer cost Everpix more money than it made from that customer, which means that growth accelerated the business failure.  It may have been possible to redesign the product so each additional customer cost less money, but without looking at the code or designs I have no idea if it were possible.  The business as implemented and the cost structures it faced were not sustainable at any proximate production frontier.  Period.  There was nothing they could have done other than shut down the company earlier and return more of the venture money.

In general higher volume allows you to amortize fixed costs over more products so the net margins become larger.  However, when the gross margins are negative there is not a single thing you can fix with volume besides producing nothing.  Put another way, losses at Everpix were positively correlated with usage and growth.

Seeing the data is super cool
None of this analysis would be possible without the Everpix team releasing their data.  Thank you to them.  I typically provide these research reports privately, but because Everpix made its data public I thought I should reciprocate.  More companies should be transparent about their deals since it is the best way for the community as a whole to thrive.

Parting words to Everpix

I wish the Everpix team the best of luck and that they learn from this experience.  Thank you so much for releasing the data for the next generation of startups to study and grow. You had a compelling product that was attractive and buzzworthy. I just hope that you take away how important variable costs are for your business and to pay attention to them in the future.

Coda

Last, if you like analyses like this, I generally author private research and presentations based on technical audits of startups.  My typical customers include VCs, angels, and sometimes startups themselves to remove any blinders and bring in an outside perspective.  As a developer who understands economics and accounting, I try to put the two together in useful ways. Contact me for more information.  My email is ivanplenty at the mighty, mighty gmail.